A harmonized dataset and exploratory non-destructive screening framework for characterizing physical and mechanical interrelationships in traditional heritage materials Authors: Mohammed A Albadrani Published: 2026 ABSTRACT AbstractCharacterizing the mechanical and thermal properties of heterogeneous natural building materials, including traditional stones and earth-based composites, presents significant engineering challenges. This study proposes an exploratory non-destructive analytical framework to examine potential proxy relationships among physical, mechanical, and hygrothermal properties of traditional heritage materials. Because the dataset consists of seven aggregated material classes, the results should be interpreted as preliminary trends rather than validated predictive laws. Using a dataset covering seven aggregated material classes, the analysis suggests that porosity may serve as an exploratory indicator of compressive strength (R² = 0.62), while density shows a strong association with thermal conductivity (R² = 0.85). These relationships should be interpreted as dataset-specific trends rather than externally validated predictive models. Furthermore, Principal Component Analysis (PCA) was used as an exploratory dimensional-reduction tool to visualize broad material groupings associated with density, porosity, moisture absorption, and strength. PC1 accounted for 82.7% of the variance within the analyzed dataset; however, this result should not be interpreted as a validated classification model because of the limited number of material classes. The present study does not implement a full HBIM or digital twin system. Instead, it proposes a conceptual pathway for future integration: (1) non-destructive measurement of porosity/density, (2) input into a material-property database, (3) assignment of material parameters to HBIM elements, (4) risk classification using PCA/clustering, and (5) visualization of vulnerable elements for conservation decision-making.Keywords: Non-Destructive Testing (NDT), Natural Building Materials, Exploratory modeling, PCA (Principal Component Analysis), Material Characterization, Structural IntegritySubject terms: Engineering, Materials science Introduction Preserving historic structures requires a delicate balance between cultural authenticity, structural stability, and sustainability under environmental and urban pressures. Traditional materials including adobe, lime mortar, natural stones (e.g., limestone, marble, sandstone), and timber are fundamental components of vernacular heritage. These materials are characterized by complex mechanical properties, significant thermal mass, and intricate pore structures. However, their engineering properties, such as compressive strength, thermal conductivity, and moisture absorption, exhibit high stochastic variability due to environmental exposure and aging processes1,2. Recent advances in heritage science and rock mechanics have introduced robust non-destructive diagnostic tools. Terrestrial Laser Scanning (TLS) and photogrammetry facilitate the precise documentation of architectural geometries3, while Historic Building Information Modeling (HBIM) enables the integration of material-specific data into digital reconstructions4,5. Furthermore, digital twins and numerical simulation approaches are increasingly applied to predict structural responses under environmental loading6,7. Specialized techniques such as thermal imaging8, energy modeling9, and nanomaterial-based conservation treatments10 have further expanded the technical toolkit available for heritage interventions. Despite these advancements, the comparative integration of physical and mechanical properties across diverse traditional materials remains underdeveloped. Prior research typically examines single lithologies (e.g., limestone or marble) or focuses on isolated property domains such as mechanical strength2, thermal behavior9, or hygric absorption11,12. In this study, fired clay brick was excluded because its industrial manufacturing process results in relatively standardized properties. In contrast, the variable and non-standardized nature of traditional materials like adobe and natural stone requires more sophisticated, multivariate diagnostic frameworks. A significant gap exists in the application of multivariate statistical methods such as regression analysis, Principal Component Analysis (PCA), and hierarchical clustering within heritage science13–15. While individual correlations between strength, porosity, and density are well established, fewer studies have organized these relationships into a transparent comparative workflow across both earthen and stone heritage materials. Therefore, the contribution of this study is integrative and exploratory rather than the discovery of new universal material laws. This study addresses this gap by developing an exploratory multi-parameter screening workflow for seven traditional construction materials (adobe brick, lime mortar, limestone, sandstone, marble, volcanic stone, and wood plank). Utilizing a synthesized dataset from twelve peer-reviewed studies (2015–2024), the specific objectives are to: Quantify correlations among compressive strength, open porosity, bulk density, thermal conductivity, and moisture absorption. Develop exploratory regression models to examine whether non-destructive indicators, particularly porosity and density, can support preliminary screening of mechanical and thermal performance. Classify materials into conservation-relevant clusters using PCA and Ward’s hierarchical clustering to facilitate risk screening. Recent developments in materials informatics have favored highly complex, data-driven frameworks. For instance, continuous localized environmental stressors like temperature can be dynamically tracked using fiber Bragg grating sensors embedded directly within structural layers to capture real-time physical anomalies. Similarly, advanced machine learning approaches have been successfully deployed to isolate carbon footprint variations across highly volatile concrete composition records, while advanced ensemble techniques, such as Levy-flight enhanced decision trees paired with SHAP analytics, have elevated the predictive accuracy of complex structural structural shear parameters. Furthermore, data-driven forecasting algorithms have proven critical for mapping long-term durability metrics, such as compressive degradation under extreme thermal exposure. However, these highly parameterized artificial intelligence configurations typically demand massive, uniform datasets (N > 102). In historical conservation projects, extensive material sampling is strictly restricted by international conservation charters. In this context, the present study illustrates how simple statistical workflows, including Pearson correlation, exploratory regression, PCA, and hierarchical clustering, can organize low-density aggregated heritage-material data into transparent preliminary screening indicators. These indicators are not proposed as validated predictive tools but as a basis for future experimental validation and HBIM-oriented data structuring. The contribution of this work lies in the systematic integration of dispersed datasets into a unified model. Unlike earlier studies that treated materials in isolation, this framework supports preliminary comparative screening and may inform future heritage risk assessment, HBIM documentation, and energy-retrofit studies after independent validation. Literature review Current scholarship highlights the fragmented nature of heritage characterization. Research on stone and earthen materials emphasizes high vulnerability to weathering; Ortega-Morales and Gaylarde11 documented how biological colonization accelerates limestone decay, while Rossi and Bournas2 demonstrated structural risks in masonry elements like Basaltic Tuff and marble. While digital twins and HBIM offer platforms for linking this data4–7, most applications remain qualitative or geometry-focused. Thermal and hygric studies, such as those by Resende et al.8 and Franco and Mauri9, confirm the roles of density and porosity in thermal behavior but often focus on case-specific retrofits. Similarly, nanotechnology and adaptive reuse studies10,12,15–19 provide context-specific responses but lack the cross-material classification established here. By applying standard engineering statistical tools to heritage datasets, this study offers a transparent exploratory workflow for preliminary conservation screening within the limits of the analyzed dataset. Recent studies increasingly combine non-destructive sensing, statistical interpretation, and machine-learning models for material characterization and structural assessment20–32. Mieloszyk et al.20 used embedded Fiber Bragg Grating sensors to evaluate temperature effects in additively manufactured structures, showing the role of sensor-based NDT in structural health monitoring. Çiftçioğlu et al.21 developed a machine-learning causal inference framework for assessing the carbon footprint of sustainable concrete mixtures, while Çiftçioğlu et al.22 also applied Levy-flight enhanced decision trees for shear-strength prediction in reinforced concrete beams. Shafighfard et al.23 used stacked machine-learning algorithms to predict the compressive strength of steel fiber-reinforced concrete exposed to elevated temperatures. Other recent studies have used interpretable ensemble learning, random forests, gradient boosting, surrogate modeling, stochastic multiscale machine learning, and uncertainty quantification for failure prediction, thermal conductivity modeling, composite property prediction, and underground rehabilitation materials24–32. Compared with these larger and more computationally intensive studies, the present work adopts Pearson correlation, exploratory regression, PCA, and hierarchical clustering because heritage conservation datasets are often small, heterogeneous, and constrained by restrictions on destructive sampling. Therefore, the proposed framework should be understood as a transparent exploratory screening workflow rather than a fully validated machine-learning prediction system. Methods Data sources and rationale To complete this study, data was sourced from secondary materials found in published articles, technical reports and peer-reviewed journals. Sources were limited to 2015–2024 to ensure data compatibility with modern digital heritage documentation standards (e.g., TLS and HBIM). Although limited direct sampling can be performed during conservation works or on low-value fragments, extensive destructive testing on original historic materials remains restricted. This constraint justifies the present study’s reliance on secondary published data. As a result, accessing reputable published sources allows systematic analysis without damaging original materials. Secondary data found in peer-reviewed literature, technical reports and academic theses was applied in a quantitative methodology for this study. The aim was to collect information comparing traditional construction materials used in heritage work, measuring their compressive strength, porosity, density, thermal conductivity and moisture absorption. Search and Selection Strategy a systematic search protocol was followed using Scopus, Web of Science, and Google Scholar with the keywords: ‘heritage materials’, ‘traditional masonry properties’, ‘porosity’, ‘thermal conductivity’, ‘compressive strength’, and ‘hygrothermal behavior’. Only sources reporting at least two measured physical/mechanical properties for the same sample set were included. Publications were screened for methodology transparency, specimen preparation, and test standard conformity. To systematically mitigate systemic bias and ensure data harmonization across the twelve peer-reviewed sources (2015–2024), a rigorous data filtering protocol was implemented. Studies were strictly included only if their reported properties conformed to standardized testing geometries and boundary conditions (e.g., ASTM C39 for stone, EN 772-1 for masonry elements) under verified oven-dry or specific moisture-equilibrium states. Potential biases may arise from combining heterogeneous datasets, including differences in laboratory loading rates, specimen dimensions, geographic origin, mineralogical composition, moisture state, and testing conditions. These sources of variability cannot be fully eliminated in a secondary-data synthesis. To explicitly quantify this uncertainty, integrated standard deviation bands (± σ) have been extracted and mapped across the properties in Tables 1 and 2 where metadata permitted. These statistical ranges reflect the aggregated laboratory scatter inherent to secondary literature synthesis. Table 1Mechanical and physical properties of traditional materials.MaterialCompressive Strength (MPa)Porosity (%)Density (kg/m³)Thermal Conductivity (W/m·K)Moisture Absorption (%)Source/ReferenceNotesAdobe Brick1.8 ± 0.4130.2 ± 3.11650 ± 450.3518.4[12]Values harmonized; exploratory trendLimestone12.0 ± 1.5018.0 ± 1.22400 ± 601.307.1[7], [11], [12]-Lime Mortar3.2 ± 0.6525.5 ± 2.81800 ± 550.8012.3[10], [12]-Wood Plank5.1 ± 0.8212.0 ± 1.4700 ± 350.1520.0[5], [12]-Sandstone9.4 ± 1.1019.5 ± 1.92250 ± 501.109.5[7], [11]-Basaltic Tuff15.2 ± 1.3514.0 ± 1.12600 ± 401.506.3[2], [7]-Marble17.0 ± 0.952.0 ± 0.32700 ± 302.001.5[2], [10]- Mechanical and physical properties of traditional materials. Table 2Summary of literature data used for exploratory modeling.MaterialPropertyReported ValueReference No.Adobe brickCompressive strength1.5–2.5 MPa[2]Porosity25–35%[12]Moisture absorption18.4%[12]Lime mortarDensity~ 1650 kg/m³[12]Moisture absorption10–15%[10]LimestoneDensity~ 2400 kg/m³[11]Porosity~ 18%[10]Thermal conductivityMoisture absorption1.3 W/m·K7.1%[7][12]MarbleCompressive strength16–18 MPa[2]Porosity~ 2%[10]Basaltic TuffDensity2500–2600 kg/m³[2]Thermal conductivity1.4–1.6 W/m·K[7]Wood plankDensity650–750 kg/m³[5]Moisture absorption18–22%[12] Summary of literature data used for exploratory modeling. Thermal conductivity Moisture absorption 1.3 W/m·K 7.1% [7] [12] The small number of material classes (N = 7) limits statistical power and increases sensitivity to individual observations. Therefore, regression, PCA, and clustering outputs are used for exploratory risk-screening and comparative interpretation only. This aggregation supports cross-material comparison but is presented as a preliminary risk-screening baseline rather than absolute predictive values, acknowledging that intra-material variability exists. Heritage and construction context of selected materials To ensure the representativeness of the dataset, seven materials were selected based on their prevalence in global architectural heritage. These include adobe, one of the oldest construction materials used in load-bearing walls across arid regions like the Middle East and North Africa; lime mortar, the primary binding agent in pre-industrial masonry from the Roman era until the early 20th century; and limestone and sandstone, which represent sedimentary rock masonry ubiquitous in European Gothic cathedrals and South Asian temples. Additionally, the study includes marble, a metamorphic rock synonymous with monumental Mediterranean and Mughal architecture; Basaltic Tuff (specifically basaltic tuff), valued for its seismic resistance in regions such as Southern Italy and Central Mexico; and wood, a transferable organic material representing timber-frame traditions from Japanese temples to Scandinavian stave churches. By incorporating these diverse materials, the study captures a broad spectrum of mechanical, hygric, and thermal behaviors, placing each within its specific cultural and architectural heritage context. This study analyzed literature and reports that contain reliable quantitative details about traditional building materials used in heritage projects. The inclusion criteria were: (1) appearance in a peer-reviewed journal or at a technical institution, (2) matching results to traditional building methods, (3) having numerous physical and mechanical properties and (4) detailed methods and test specimens. Twelve key references published over 2015 to 2024 were reviewed and each focused on regions and climates that varied greatly. Materials and variables The following materials were selected for analysis: Due to their porosity and good thermal mass, adobe bricks help regulate indoor temperatures in hot and dry areas7,12. Due to their porosity and good thermal mass, adobe bricks help regulate indoor temperatures in hot and dry areas7,12. Limestone and sandstone were included because they are widely used in traditional and classical stone masonry8,11. Limestone and sandstone were included because they are widely used in traditional and classical stone masonry8,11. Lime mortar which belongs to the traditional group and has hygroscopic features9. Lime mortar which belongs to the traditional group and has hygroscopic features9. Wood plank is frequently seen in both support structures parts5. Wood plank is frequently seen in both support structures parts5. Most monuments built with Basaltic Tuff and marble are strong and not porous2,10. Most monuments built with Basaltic Tuff and marble are strong and not porous2,10. Because the source studies reported data using different formats, including ranges, means, and representative values, all values were harmonized into SI units and converted into material-level central estimates. When ranges were reported, mid-point or reported mean values were used, and the available variability was retained as uncertainty ranges where possible. This harmonization enables cross-material comparison but may introduce bias due to differences in specimen age, origin, preparation, moisture condition, testing standard, and laboratory procedure. Accordingly, the dataset should not be interpreted as a substitute for controlled experimental testing. It provides a comparative baseline for heritage material screening, while intra-material variability remains an important source of uncertainty. This approach supports cross-material comparison but does not capture intra-material mineralogical or regional variability, which is acknowledged as a limitation. Testing standards and property definitions All material properties were extracted exclusively from studies reporting standardized laboratory testing procedures. For clarity and comparability, the following definitions and standards were adopted: Compressive strength: determined according to ASTM C39 (stone materials), ASTM C67 (masonry units), or EN 772-1 for masonry specimens, depending on material type. Bulk density (apparent density) was used consistently across all materials, determined under oven-dry conditions according to ASTM C97 (stone) or EN 772 − 13 (masonry). Porosity refers to open porosity (%), calculated from bulk and true density or by water saturation methods following ASTM C97 or EN 1936. Moisture absorption (%) refers to water absorption at full saturation (constant mass under atmospheric pressure), unless explicitly specified in the source study. Fixed-time immersion results were excluded to ensure comparability. Thermal conductivity values were obtained using steady-state guarded hot plate or heat flow meter methods according to ASTM C177 or ISO 8302. To ensure comparability across different lithologies, open porosity (%) was calculated using water saturation methods (ASTM C97), and bulk density (kg/m³) was determined under oven-dry conditions. This distinction is critical for understanding the hygrothermal behavior of porous materials like adobe versus dense rocks like marble. Where different but equivalent international standards were used in the source studies, equivalence was verified prior to inclusion. All materials were included in a single row, sharing the average values reported in scientific articles. SI units were used to standardize these values, and a table (Table 1) was created so the values can be checked against a single list. Moisture absorption (%) refers to the mass gain under free water saturation at atmospheric pressure, unless otherwise specified in the source study. Moisture absorption (%) refers to mass gain under full water saturation (constant mass at atmospheric pressure), representing open porosity–accessible water absorption. Studies reporting partial immersion or fixed-time absorption were excluded to ensure inter-material comparability. To perform advanced statistical operations, the dataset was imported into Minitab 21. Pearson correlation was applied to evaluate linear relationships among variables. Simple linear regression was used to explore dataset-specific associations between independent variables (porosity, density) and dependent variables (compressive strength, thermal conductivity). All variables were standardized prior to PCA using z-scores. PCA extraction followed the Kaiser criterion (eigenvalues > 1) and loading interpretation thresholds of |0.40|. Hierarchical clustering was conducted using Ward’s method with Euclidean distance to group materials based on the similarity of their physical and mechanical profiles, identifying coherent clusters consistent with physical understanding. PCA was employed to reduce dimensionality and retain the most significant components, with selection based on eigenvalues greater than 1 and cumulative variance explained. A significance level of α = 0.05 was reported for descriptive statistical interpretation only. Given the limited number of material classes, p-values were treated cautiously and were not used as evidence of external predictive validity. Several statistical techniques were applied to evaluate relationships among the selected material properties. First, several measures such as mean, minimum, maximum and range, were used to explain the activity of each variable. To identify important links between porosity, density and strength, Pearson’s correlation coefficients were used. Then, Regression models were used to explore whether porosity and density may serve as potential proxy indicators. Due to the limited sample size, leave-one-out cross-validation (LOOCV) was used only as an internal sensitivity check for the two exploratory regression models. Because N = 7, the LOOCV results should not be interpreted as external validation. They are reported to show how sensitive the fitted relationships are to the removal of individual material classes. No claim of external generalizability is made. The proposed equations should be used only within the range of the analyzed dataset and should be validated using independent experimental or in-situ monitoring data before practical application. To uncover more about underlying patterns in the data, PCA was used to shrink the number of dimensions, allowing materials to be organized by their resemblance in a smaller set of features. Another technique included in this study is hierarchical cluster analysis which organized materials into groups based on their performance, helping to spot where they are alike or different. The results were explained with scree plots, biplots, dendrograms and loading plots, all of which were produced using α = 0.05 as the confidence level. Table 2 summarizes the literature sources used for deriving average values of material properties. These references provide the foundation for the statistical analyses presented in Tables 2, 3, 4, 5, 6, 7, 8 and 9. Table 3Pearson correlation matrix among key material properties.Compressive Strength (MPa)Porosity (%)Density (kg/m³)Thermal Conductivity (W/m·K)Porosity (%)−0.7880.035Density (kg/m³)0.777−0.2880.0400.532Thermal Conductivity (W/m·K)0.914−0.6130.9250.0040.1440.003Moisture Absorption (%)−0.8960.577−0.928−0.9950.0060.1750.0030.000 Pearson correlation matrix among key material properties. Table 4Analysis of variance for exploratory regression model: compressive strength vs. porosity.SourceDFAdj SSAdj MSF-ValueP-ValueRegression1131.67131.678.170.035Porosity (%)1131.67131.678.170.035Error580.5516.11--Total6212.22--- Analysis of variance for exploratory regression model: compressive strength vs. porosity. Table 5Analysis of variance for exploratory regression model: thermal conductivity vs. density.SourceDFAdj SSAdj MSF-ValueP-ValueRegression12.16292.1629129.520.003Density (kg/m³)12.16292.1629129.520.003Error50.36640.07327--Total62.5293--- Analysis of variance for exploratory regression model: thermal conductivity vs. density. Table 6Statistical relationships between material properties (Provides a concise interpretation of the regression models, R-squared values, and observed trends).Variable 1Variable 2Observed relationshipS (Std. Error)R²Adjusted R²Predicted R²NotesPorosity (%)Compressive Strength (MPa)Negative correlation (r ≈ −0.788)4.0162.05%54.46%42.15%Exploratory trend; dataset-specificDensity (kg/m³)Thermal Conductivity (W/m·K)Strong positive linear trend (r ≈ 0.925)0.92585.51%82.62%43.37%Dataset-specific; caution for generalization Statistical relationships between material properties (Provides a concise interpretation of the regression models, R-squared values, and observed trends). Table 7Principal component analysis – Eigenvalues and variance.Eigenvalue4.13410.79070.06440.01000.0008Proportion0.8270.1580.0130.0020.000Cumulative0.8270.9850.9981.0001.000 Principal component analysis – Eigenvalues and variance. Table 8PCA Eigenvectors and variable loadings.VariablePC1PC2PC3PC4PC5Compressive Strength (MPa)0.474−0.203−0.778−0.316−0.173Porosity (%)−0.3430.803−0.221−0.335−0.277Density (kg/m³)0.4350.517−0.1940.5610.437Thermal Conductivity (W/m·K)0.4860.1290.3440.217−0.763Moisture Absorption (%)−0.482−0.173−0.4360.653−0.348 PCA Eigenvectors and variable loadings. Table 9Hierarchical cluster analysis – amalgamation steps.StepNumber of active clustersSimilarity Level (%)Distance levelFirst Material/Cluster JoinedSecond Material/Cluster JoinedResulting clusterNotes1699.7510,171Basaltic TuffMarbleHigh-Density Crystalline Rock GroupExploratory cluster2599.4422,515LimestoneSandstoneSedimentary Masonry Unit Group-3499.4322,561Adobe BrickLime MortarPorous Earth-Based Composite Group-4394.71211,579Sedimentary Group (Limestone/Sandstone)Crystalline Group (Basaltic Tuff/Marble)Consolidated Dense Natural Stone Matrix-5264.981,401,206Porous Earth Group (Adobe/Lime Mortar)Wood PlankConsolidated Lightweight Hygroscopic MediaCorrected cluster order per reviewer comment61−4.514,181,098Consolidated Stone MatrixConsolidated Hygroscopic MediaUnified Heritage Material Macro-ArrayNegative similarity noted; Ward’s method artifact Hierarchical cluster analysis – amalgamation steps. Results This section presents the results of comprehensive statistical and engineering analysis of traditional building materials used in heritage structures, combining both mechanical and physical properties sourced from diverse peer-reviewed literature and experimental reports. The exploratory cross-correlation framework evaluated across the macro-averaged material groups revealed a moderate statistical relationship between open porosity and compressive strength (R² = 0.6205, p = 0.035) and a prominent relationship between bulk density and thermal conductivity (R²= 0.8551, p = 0.003). Due to the restricted sample size (N = 7), these mathematical equations must be interpreted strictly as exploratory statistical trends matrices rather than generalized, absolute predictive property laws. A critical limitation of utilizing macro-aggregated mean values from secondary literature is that it inherently masks intra-material stochastic variability. Micro-level heterogeneities, including localized micro-cracking, variable mineral distribution, anisotropic pore networks, and historical aging artifacts, are compressed into single data vectors. Consequently, these models serve as an evaluative baseline workflow for preliminary risk screening rather than high-fidelity micro-mechanical forecasting tools. Descriptive properties and ranges Table 1 clearly indicates that traditional heritage materials differ widely in their physical and mechanical properties. For example, compressive strength ranges from 1.8 MPa in adobe brick to 17.0 MPa in marble, this range illustrates which materials primarily function as load-bearing elements (e.g., marble, volcanic stone) compared with those used for infill or thermal regulation (e.g., adobe, lime mortar). Just as with permeability, earthen materials like adobe have much greater porosity, for instance 30%, than 2.0% in marble, illuminating their greater insulating capabilities than stones. Density results agree, placing wood plank as the lightest (700 kg/m³) and marble as the heaviest (2700 kg/m³). Such patterns show that each material has unique strengths and weaknesses in strength, weight and being thermally active. Correlation analysis According to the results in Table 3 shows statistically significant associations among selected physical and mechanical properties. Open porosity was negatively correlated with compressive strength (r = − 0.788, p = 0.035), indicating that, within this dataset, more porous materials tended to exhibit lower strength. Bulk density was strongly and positively correlated with thermal conductivity (R = 0.925, p = 0.003), suggesting that denser materials such as marble and Basaltic Tuff had higher heat-transfer capacity. Moisture absorption was negatively associated with density and thermal conductivity, reflecting the contrasting behavior of porous and dense material groups. Regression models and residual diagnostics. In Fig. 1, the residual diagnostic plots provide a visual check of the exploratory regression model relating density to thermal conductivity. Because the dataset contains only seven material classes, these plots should be interpreted cautiously and cannot confirm distributional assumptions with high statistical confidence. The residuals do not show an obvious systematic trend, but the small sample size limits the reliability of formal residual-based inference. Therefore, the diagnostic plots are reported only as a supporting visual assessment, not as evidence of model validation. Residual diagnostic plots for the exploratory regression model of thermal conductivity versus density: (a) Normal probability plot of standardized residuals; (b) standardized residuals versus fitted values; (c) histogram of standardized residuals; and (d) standardized residuals versus observation order. Running regression analysis also supports these relationships. Within the analyzed dataset, the exploratory regression between open porosity and compressive strength was expressed as follows Eq. (1): This equation is dataset-specific and should not be applied as a general predictive law without independent validation. In situations where destroying samples is not allowed, checking porosity is an easier method for diagnostics than destructively testing compressive strength. Within the analyzed dataset, the exploratory regression between density and thermal conductivity was expressed as follows Eq. (2): This relationship should be interpreted as a dataset-specific screening trend rather than an externally validated prediction model. Within the analyzed dataset, the observed density–thermal conductivity trend suggests that density may be a useful screening indicator when assessing thermal behavior of similar traditional materials. The observed density–thermal conductivity relationship is based on harmonized values and may change under varying moisture conditions. In porous materials, increased moisture content can increase effective thermal conductivity because water-filled pores transfer heat more efficiently than air-filled pores. However, practical retrofit decisions require project-specific validation. (see also Table 6). Two primary statistical relationships were identified within the analyzed dataset: the negative association between porosity and compressive strength, and the positive association between density and thermal conductivity. Moisture absorption was interpreted qualitatively as an indicator of potential moisture vulnerability, particularly for adobe and lime mortar. However, no independent decay-rate variable was included in the dataset; therefore, no regression model was fitted for material decay. To evaluate the internal sensitivity of the exploratory regression models, Leave-One-Out Cross-Validation (LOOCV) was applied. For the porosity–compressive strength model, the predicted R²/LOOCV-based sensitivity metric was 42.15%, compared with a sample R² of 62.05%. For the density–thermal conductivity model, the predicted R²/LOOCV-based sensitivity metric was 43.37%, compared with a sample R² of 85.51%. These reductions indicate limited cross-validated stability and confirm that the fitted relationships are sensitive to individual material classes. Therefore, the models should be interpreted only as exploratory screening equations and not as validated predictive tools. Principal component analysis Underlying patterns were discovered among a number of variables by means of Principal Component Analysis (PCA). As demonstrated in Table 7, PC1 accounts for 82.7% of the variation by showing strong positive associations with compressive strength, density and thermal conductivity and negative associations with porosity and the amount of water absorbed. Therefore, PC1 can be interpreted cautiously as a dataset-specific contrast between dense, low-porosity materials and lightweight, moisture-sensitive materials. Because PCA was applied to only seven material classes, this component should be treated as an exploratory visualization axis rather than a definitive material-classification rule. Hierarchical clustering (Table 9) produced two primary groups: (1) porous, lightweight, moisture-sensitive materials (adobe, wood, lime mortar), and (2) dense, low-porosity, high-strength stones (marble, volcanic stone, limestone). The amalgamation steps correspond to their progressive joining based on similarity distances: Cluster 1 (adobe + wood) with low strength, high porosity, and moisture sensitivity, and Cluster 2 (marble + volcanic stone) with high density, strength, and low porosity. The clustering analysis grouped materials according to similarity in their standardized physical, mechanical, and hygrothermal profiles. Clustering steps show that adobe and wood are put together first, since they have similar properties of low strength and high moisture sensitivity, while marble, Basaltic Tuff and limestone gather later for their similar features of high strength and less porosity. These clustering patterns are consistent with the PCA results and with the expected physical distinction between porous, moisture-sensitive materials and dense, low-porosity stone materials. The results of correlation, regression, PCA, and clustering are presented in Tables 2, 3, 4, 5, 6, 7, 8 and 9. A strong correlation between density and thermal conductivity indicates that denser materials, such as marble and limestone, transfer heat more readily. This aligns with broader work on integrating renewable technologies in heritage buildings, as reviewed by Cabeza et al.19. This has direct implications for retrofitting and energy performance in historic buildings. Similarly, porosity shows an inverse relationship with compressive strength, confirming its role as a practical, non-destructive indicator of structural performance. In clustering results (Table 9), adobe and wood grouped together due to their high porosity and moisture sensitivity, whereas marble and Basaltic Tuff clustered together because of their high density and durability. These practical insights provide useful guidelines for conservation decisions. This 3D surface plot provides a vivid visualization of how material porosity (%) and density (kg/m³) interact to influence a critical mechanical property: compressive strength (MPa). In the context of traditional materials used in historical heritage structures, such as mudbrick, limestone, tuff, or adobe, this type of analysis is essential for understanding their long-term performance and structural integrity as presented in Fig. 2. 3D surface plot of material properties. Due to the limited number of material classes (N = 7), regression results should be interpreted as exploratory screening trends rather than transferable laws. Adjusted and predicted R² values are therefore reported to avoid overestimation of explanatory power. To provide a holistic comparison of all materials across their normalized mechanical and physical properties, a radar plot was generated (Fig. 3). The visualization highlights the contrasting performance profiles: adobe and lime mortar exhibit high porosity and moisture absorption with low strength, whereas marble and Basaltic Tuff demonstrate high strength and density with low porosity. Such multi-property visualization reinforces the statistical outcomes (Tables 3, 4, 5 and 6) and offers a practical tool for conservation engineers to quickly assess material vulnerabilities and trade-offs. Radar plot comparing normalized properties (strength, porosity, density, thermal conductivity, moisture absorption) of all seven heritage materials. Discussion The statistical integration of mechanical, thermal, and moisture-related properties across seven aggregated traditional material classes highlights dataset-specific patterns that are consistent with established material behavior. As expected, porous materials such as adobe and lime mortar exhibit low compressive strength and high moisture absorption, confirming their well-documented vulnerability in arid and semi-arid climates2,12. Open porosity can reduce compressive strength by decreasing the effective load-bearing solid matrix, increasing local stress concentration around pores, and facilitating moisture ingress that may accelerate micro-cracking and material degradation. Conversely, marble and Basaltic Tuff demonstrate high strength and density with low porosity, consistent with their durability but also with their higher thermal conductivity. These findings are consistent with earlier studies on material decay and stone mechanics2,11 and extend them by organizing property relationships across different materials within a transparent exploratory framework rather than treating them in isolation. The correlation between density, porosity, and thermal transport observed in these natural materials mirrors the behavior of other anisotropic bio-based materials, such as wood and natural fiber composites. This suggests that the exploratory PCA-based grouping may provide a useful basis for future comparative studies of porous natural media, provided that larger and independently validated datasets are used. The classification of marble and Basaltic Tuff into a single cluster is justified by their low open porosity (< 14%) and high bulk density (> 2600 kg/m³) which results in superior mechanical resilience but higher thermal conductivity. This contrast with the high-porosity Group 1 (adobe, wood, mortar) highlights the necessity of differentiated conservation strategies. Nevertheless, the study is limited by its reliance on secondary datasets. Future work should test these exploratory relationships against long-term performance data and expand them to include properties such as salt crystallization resistance. Additionally, intelligent detection approaches for stone deterioration have been explored in recent research, such as the model developed by Hatır et al.18. By applying regression, PCA (Principal Component Analysis), and clustering tools common in engineering but underused in heritage science, this study demonstrates how dispersed experimental data can be synthesized into an exploratory comparative framework for preliminary conservation screening. PCA provided a preliminary visualization of the contrast between dense, low-porosity materials and lightweight, moisture-sensitive materials within the analyzed dataset, while hierarchical clustering confirmed these groupings. This comparative classification may support preliminary material grouping and risk-screening discussions when destructive sampling is restricted. However, it should be used only as a decision-support aid and not as a stand-alone diagnostic or risk-detection system. The correlations between porosity strength and density thermal conductivity have been recognized individually, but their systematic integration into an exploratory screening workflow may support preliminary conservation planning after independent validation. For instance, porosity may serve as a preliminary non-destructive indicator of compressive strength within the analyzed dataset. To clarify how this exploratory framework interfaces practically with digital conservation platforms, a conceptual integration pipeline is proposed for Historic Building Information Modeling (HBIM) architectures, framed here as a foundational workflow for future development in Fig. 4. Conceptual workflow for integrating non-destructive material indicators into HBIM-based heritage risk screening. In the proposed conceptual pipeline, non-destructive measurements such as ultrasonic pulse velocity, infrared thermography, micro-CT-derived porosity, or localized moisture measurements could be used as input fields for future HBIM-linked material databases. These values could then be processed through externally validated empirical relationships or scripts and assigned to HBIM material parameters. These parameters could be stored as HBIM material attributes or IFC-compatible property sets after software implementation and field validation. In the present study, this workflow is not implemented in a live HBIM platform; it is proposed as a future integration pathway that requires software implementation, field calibration, and validation against in-situ measurements. If implemented and validated in future studies, such parameter mapping could enrich HBIM elements with material-related information for conservation decision support. Porosity can be estimated non-destructively using several indirect techniques, including micro-CT scanning, ultrasonic pulse velocity calibrated with empirical porosity models, infrared thermography paired with inverse heat-transfer modeling, and NMR-based surface moisture diffusion measurements. These methods provide practical options when destructive sampling is prohibited. Nevertheless, the study is limited by its reliance on secondary datasets. While this approach reflects the ethical and regulatory constraints of sampling historic materials, it also highlights the need for validation through in-situ monitoring and experimental testing. Future work should test these exploratory relationships against long-term performance data and expand them to include properties such as salt crystallization resistance and freeze–thaw durability. This study contributes to heritage science by organizing existing material-property relationships into an exploratory multi-parameter screening workflow. Preliminary non-destructive indicators: porosity and density may support initial screening of compressive and thermal behavior, provided that project-specific validation is performed. Comparative classification: PCA and clustering distinguish vulnerable, porous materials (e.g., adobe, wood, lime mortar) from dense, durable ones (e.g., marble, volcanic stone, limestone). Potential future applications: after validation with larger experimental and in-situ datasets, the proposed workflow could support conservation risk screening, energy-retrofit decision-making, regional heritage initiatives such as Saudi Vision 2030 and HBIM-linked material documentation. Future sensitivity analysis and physics-informed modeling Because the present dataset contains only seven aggregated material classes, a full quantitative global sensitivity analysis was not performed. Nevertheless, future work should examine how uncertainty in porosity, density, moisture content, anisotropy, and mineralogical composition propagates into predicted mechanical and thermal properties. This could be achieved using one-at-a-time sensitivity analysis, Monte Carlo uncertainty propagation, or variance-based Sobol indices once larger datasets become available. A further research direction is the integration of physics-informed neural networks (PINNs) or hybrid solver-based models. In such a framework, empirical material-property data could be combined with governing equations for heat transfer, moisture diffusion, and mechanical degradation. From a solver perspective, finite-element or hygrothermal simulation outputs could be used to calibrate or constrain machine-learning models, allowing predictions to remain consistent with physical conservation laws. However, such approaches require substantially larger, site-specific, and time-dependent datasets than those available in the present study. Therefore, PINN-based digital twins and real-time monitoring are proposed only as future developments, not as demonstrated outcomes of this work. Conclusion This study developed an exploratory multi-parameter framework for organizing aggregated secondary data on seven traditional heritage material classes. The results indicate dataset-specific associations among compressive strength, open porosity, density, thermal conductivity, and moisture absorption. Within the analyzed dataset, porosity showed a moderate negative association with compressive strength, while density showed a strong positive association with thermal conductivity. PCA and hierarchical clustering further separated porous, moisture-sensitive materials from denser stone materials. Due to small sample size (N = 7) and reliance on secondary data, results should be interpreted as preliminary screening trends rather than generalized predictive laws. These findings should be interpreted as preliminary screening trends rather than externally validated predictive laws. The small sample size, use of aggregated material-level values, and reliance on heterogeneous secondary sources limit statistical generalization. Therefore, the proposed equations and classifications require validation using independent experimental measurements, in-situ monitoring, and larger material datasets before practical deployment. The main contribution of the study is the integration of existing material-property relationships into a transparent exploratory workflow that can support future HBIM-based heritage risk screening. Future research should expand the dataset, include regional variants, evaluate uncertainty more comprehensively, and incorporate additional durability indicators such as salt crystallization resistance and freeze–thaw behavior. Non-destructive measurements (porosity, density) require field validation before HBIM integration or use in conservation decision-making. Acknowledgements The Researchers would like to thank the Deanship of Graduate Studies and Scientific Research at Qassim University for financial support (QU-APC-2026). Author contributions Mohammed A. Albadrani: Conceptualization; Methodology; Data curation; Formal analysis; Visualization; Writing – original draft; Writing – review & editing; Supervision; Funding acquisition. Data availability The data sets used during the current study are available from the corresponding author on reasonable request (due to privacy). Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Citations Data Availability Statement The data sets used during the current study are available from the corresponding author on reasonable request (due to privacy).